I've built something very similar but am further along (https://github.com/stravu/crystal). I understand the desire to do your own thing, but if you are interested in joining forces and contributing I would love to have you. I think we were thinking along very similar lines.
I tried your project in my exploration and ran into an issue but I can't quite remember what it was. Anyways, I would be interested in connecting and learning more about your project, I dropped my X handle in the first issue on my Github repo
Take a look at https://conductor.build/ - they really pioneered this whole direction of using git worktrees. Cursor will soon look like their app, Opencode will soon look like their app, etc.
I tried that, I couldn't get it to authenticate properly since I need to run some commands before running claude code. I also don't like that they rolled an entire UI when the terminal UI is good enough, it's just the parallelization of workflows that needed automation. Git worktrees were launched in 2015
I am a Kanban person and I use Trello. I couldn't find a Trello MCP that worked so I built https://github.com/billyjones75/cueit which is basically a Kanban board for coding agents.
Validating to see projects like these coming to life. I'd also love your feedback on our take of this development approach! We started https://devswarm.ai about 6 months ago and we're aiming to solve this problem in similar ways!
* Agent UI is the CLI itself, no wrapper UI.
* Isolate your work with worktrees (though we do our best to make sure you never have to think about them!)
* Many agents to choose from (Claude, Codex, Gemini, Amp, etc), and more to come!
* Features to support running your app in parallel (port vars, untracked file copying).
* Easy access to your local IDE from the worktree.
* Soft launch of our Jira integration if you click on the repository name in the sidebar and click “Jira”, start your agents quickly right from a backlog-esque view of your work! (Next release will integrate Jira into onboarding)
Feedback on your app, I love that I can see the connectivity of the MCP servers right there. I always get jarred when Atlassian disconnects and suddenly my AI wants to hit the Jira REST API for some reason.
I had the same issues, and half-baked a similar solution. But then I looked into dev containers, with which I get higher isolation, including a DB for each instance (which is important for testing in parallel; I'm mostly using Ruby on Rails). What I'm doing now is:
* create a dev container for the project
* install the agent (Claude Code in my case) in the container as part of the dev container definition
* launch the container through DevPod (no affiliation) which automatically connects VS code or a JetBrains agent
So now I can run these in parallel, on a remote server if I want, and in "YOLO" mode. Personally, I'm finding this superior to the git worktree alternative.
Hi OP, can you tell me how you got started to build tool like this? From what I understand, you are leveraging git worktree under the hood and using LLM agent that work in CLI to maintain 'branch' of version being produced. depending on which agent produces the best version of code, you pivot to that one. And Fleetcode provider solution to manage this.
Did you use TypeScript to build FleetCode because of electron app?
For anyone who either doesn't want a GUI-based workflow or wants a lot of custom behavior, I would encourage you to roll your own tool, either by yourself or with an agent. I did this at work and the tool became very usable in short order, building on itself in a virtuous cycle with built-in handling for the quirks of our engineering org and my own preferences. The codebase I ended up with was quite ugly, but the problem space of doing this for local dev is small enough to be manageable.
It does git worktree based parallelisation as well.
Edit: Ah! I see you mentioned several tools. Sorry I saw the repo and immediately thought of container use as I have been planning to give it a go this week. Did you happen to try it too? And where did it breakdown?
I'll take a look. I think containers are overkill for my type of workflow. I'm working on a project where dependencies don't change often so I'm not worried about a CLI agent installing something on my machine and breaking things.
This is cool, except I don’t want to run multiple copies of my dev stack.
GitButler does this cool thing where you can work on multiple branches at the same time, applied to the same working directory. For example you can work on the css while an agent works on the admin panel on a separate branch. It would be cool to have this with a tool like FleetCode.
I couldn't figure out this part of GitButler. I downloaded it and tried it (since I respect Scott Chacon and remember learning so much about Git from his well-written git guides).
But it was just a bit too much cognitive dissonance for me to try it.
It seems like the next big thing is parallel coding... I've tried GitButler, Spectator, Vibe-Kanban, and Conductor in the past week. And there is now FleetCode.
I liked Spectator's idea (use a separate docker container for each) but it didn't quite work right. So back to worktrees which seem to work just fine.
At some point, we will probably consolidate on 2-3 dominant tools in this space.
I wonder if even work trees will be needed if we can do a "create a copy-on-write version of my code folder" which would result in nearly zero-cost copies of the repo.
Somehow the GitButler workflow works great for me and it is the first VC software that made me drop Magit after a decade of daily use.
I do not use it collaboratively. I use it to continuously ship smaller things while working on bigger pieces and I constantly move independent changes around to different “lanes” to ship frequently as parts of my work mature.
With Magit I used staging area and amended commits continuously. With GitButler I “assign” files or chunks by dragging them into lanes as I am happy with the changes, and when I have a logical unit I commit it. Having this multiple staging areas has been a great workflow improvement as well.
can you elaborate why you don't want to run multiple copies of your dev stack? is it because you want to run your app from one place but be able to test multiple changes, vs having to install deps and start it from multiple folders?
Former. Otherwise I would need to have multiple database instances running, each with its own data and migrations to keep in sync. Plus I would need to re-do local env vars to hook all of this up for each worktree etc. And I often want to know how these branches play together, with GitButler I can bring these branches in and out with a single “apply to workspace” click.
Is this based on a single provider? Could you explain how it differentiates itself from competing solutions? Perhaps an end-to-end workflow walkthrough or a short video would be helpful to understand the process before attempting to implement it. Thank you for open-sourcing the code.
Not based on a single provider, I need to add more providers but basically supporting a provider is as simple as adding the relevant shell command for the CLI agent(claude for CC, codex for OpenAI etc). I'll try making a video at some point.
I'm not trolling at all, I genuinely liked that it's not some pay-to-use, SaaS, startup or whatever, but an actual simple and functional tool, free, open source, documented.
Then we wouldn't need our Kanban board AND a separate board for just that.
Or I guess if these tools integrated into JIRA and suggested tickets, that could be nice.
* Agent UI is the CLI itself, no wrapper UI.
* Isolate your work with worktrees (though we do our best to make sure you never have to think about them!)
* Many agents to choose from (Claude, Codex, Gemini, Amp, etc), and more to come!
* Features to support running your app in parallel (port vars, untracked file copying).
* Basics like code editor, git commit UI, branch diff UI.
* Easy access to your local IDE from the worktree.
* Soft launch of our Jira integration if you click on the repository name in the sidebar and click “Jira”, start your agents quickly right from a backlog-esque view of your work! (Next release will integrate Jira into onboarding)
Say hi on our discord too! https://discord.gg/devswarm
We’ve been building DevSwarm with DevSwarm and building like this is just unreal. Honestly can't imagine building any other way now.
edit: Oh forgot to mention we're in free-beta right now, so please download it and tell us what you think!
Do let us know if you tried out DevSwarm.ai and if you found something missing. We built it for the same reason.
And kudos to any of us on this thread that see where coding is going! We call it high velocity engineering, or hive coding, as the code matters.
* create a dev container for the project
* install the agent (Claude Code in my case) in the container as part of the dev container definition
* launch the container through DevPod (no affiliation) which automatically connects VS code or a JetBrains agent
So now I can run these in parallel, on a remote server if I want, and in "YOLO" mode. Personally, I'm finding this superior to the git worktree alternative.
It is based off Shopify CEO Tobi's `try` implementation[1].
[0]: https://github.com/aperoc/toolkami [1]: https://github.com/tobi/try
https://github.com/stravu/crystal
https://github.com/imbue-ai/sculptor
https://github.com/omnara-ai/omnara
Did you use TypeScript to build FleetCode because of electron app?
Dagger has a nice solution for this in this space called Container Use - https://container-use.com/introduction
It does git worktree based parallelisation as well.
Edit: Ah! I see you mentioned several tools. Sorry I saw the repo and immediately thought of container use as I have been planning to give it a go this week. Did you happen to try it too? And where did it breakdown?
GitButler does this cool thing where you can work on multiple branches at the same time, applied to the same working directory. For example you can work on the css while an agent works on the admin panel on a separate branch. It would be cool to have this with a tool like FleetCode.
But it was just a bit too much cognitive dissonance for me to try it.
It seems like the next big thing is parallel coding... I've tried GitButler, Spectator, Vibe-Kanban, and Conductor in the past week. And there is now FleetCode.
I liked Spectator's idea (use a separate docker container for each) but it didn't quite work right. So back to worktrees which seem to work just fine.
At some point, we will probably consolidate on 2-3 dominant tools in this space.
I wonder if even work trees will be needed if we can do a "create a copy-on-write version of my code folder" which would result in nearly zero-cost copies of the repo.
I do not use it collaboratively. I use it to continuously ship smaller things while working on bigger pieces and I constantly move independent changes around to different “lanes” to ship frequently as parts of my work mature.
With Magit I used staging area and amended commits continuously. With GitButler I “assign” files or chunks by dragging them into lanes as I am happy with the changes, and when I have a logical unit I commit it. Having this multiple staging areas has been a great workflow improvement as well.